1. Field of the Invention
The present invention relates to a binarization method for binarizing an input original image.
2. Description of the Related Art
In recent years, color document images are increasingly used due to an improvement in capacity of computers, an increase in memory capacity, and a development of color scanner devices.
Meanwhile, character portions of a captured document image can be character-recognized to be converted into text data. In addition, in performing scanning of a large amount of originals with an automatic document feeder (ADF), an orientation of characters in each original can be determined so as to automatically determine a correct orientation of the original. The original then can be printed out in a correct orientation. In this regard, what is significant in the above-described processing is the accuracy of character recognition.
In order to improve the accuracy of character recognition, it is significant to obtain a binarized image suitable for character recognition. For example, in a binarization method discussed in Japanese Patent Application Laid-Open No. 08-223409, a type of an input image is determined, and if a result of the determination indicates an image having a single density object, the input image is subjected to simple binarization processing, and on the other hand, if a result of the determination does not indicate an image having a single density object, the input image is subjected to pseudo halftone binarization processing. In addition, in a binarization method discussed in U.S. Pat. No. 5,956,421, in performing binarization processing in units of blocks having a given size, a luminance frequency is computed based on a multilevel image in each block, and a threshold value is computed based on an average value and a skew value of the luminance frequencies. Additionally, U.S. patent application Publication No. US 2005/0047660 A1 discusses a method for binarizing reversed characters (outline characters).
However, in the conventional methods as discussed in Japanese Patent Application Laid-Open No. 08-223409 and U.S. Pat. No. 5,956,421, separate processing needs to be performed in performing character recognition on reversed characters.
Furthermore, in the binarization method discussed in U.S. patent application Publication No. US 2005/0047660 A1, partial binarization processing is performed on an image portion (area) that is determined to be a character area based on an edge image, and partial binarization processing is further performed on an image portion (area) that is determined to be a character area based on a binary image generated by simple binarization processing of a whole image. Then, an output image is generated by merging two partially-binarized images generated by partial binarization processing. Therefore, this binarization method requires processing time for performing the above-described operations.